Attention alters the way that incoming sensory information is processed in order to optimize behavior. Determining how attention influences the way that information is represented throughout the brain is a core problem in sensory neuroscience. This issue has practical consequences as well, because attention-related disorders can severely affect learning and behavior. We propose to investigate how attention changes tuning for a broad range of structural and semantic features across the entire human brain. To address this issues we propose to use functional MRI to record blood-oxygen level-dependent (BOLD) signals while humans search for specific object or action categories (e.g., Humans or Vehicles) in natural movies. We will then use an innovative voxel-wise modeling framework to characterize how attention changes tuning for hundreds of structural and semantic features. This framework will allow us to separately characterize attentional influences on response baseline, response gain and tuning. The proposal consists of two broad Aims. Aim 1 focuses on how attention changes the way that objects and actions are represented in the brain. Aim 2 addresses the computational principles that govern attentional tuning changes. These experiments will provide crucial information for understanding how the brain dynamically changes representations to optimize behavior during natural vision, and it will provide the first quantitative structural and semantic models that accurately predict BOLD responses during natural visual search.